Skip to main content
A skill is a written procedure the server can hand a model when a question matches it — which endpoints to call in which order, how to weigh conflicting signals, and how to format the result. It is loaded on demand rather than sitting in every prompt. The server exposes named tools with typed schemas; clients can defer loading or use native tool search where supported. Ask a concrete question; the agent can load the matching workflow on demand. The shared workflow bundle has five entrypoints, with deeper specialist guidance available as needed: The four Solutions workflows connect these steps to real REST requests. Ordinary MCP resource access and tool-only resource fallbacks use the same packaged content. Skill import is supported where the client offers it; an imported skill is a release snapshot and must be reimported for updates. Preparing an import bundle does not imply app-directory publication. The public coding plugin packages a snapshot of the same five named-tool workflows. Existing installs need an update and a fresh session, and a snapshot can lag the server, so compare the installed bundle manifest’s version with the server’s. Existing entrypoints such as gdelt-discover-and-drill, brief-output, multi-surface-synthesis and macro-native remain available for compatibility and specialist guidance. The prediction-market skill is retired.

The one that matters most

gdelt-discover-and-drill encodes the habit this API rewards: choose the smallest useful workflow, then refine it from returned evidence. An empty result applies only to its stated scope and coverage; missing evidence remains unknown. Reuse familiar schemas and fetch scoped options only when needed. The same loop written out by hand is in Answer a question with MCP. The brief-output skill remains available for formatting your own cited answer. Hosted Monitoring Brief generation is temporarily unavailable.

External tools

The third-party providers some skills reach for, and what they are and are not.